Spain's insect-eye sensor takes a slow path to robots that act in the real world

Most of the data a conventional camera produces is throwaway. A static wall, an empty hallway, a held pose: they all get re-captured 60 times a second, and a robot that only needs to know what moved pays the full processing cost for every pixel it never uses. A new class of event-driven camera flips that math by reporting only the pixels that changed, and it is what Ikerlan, a Spanish-Basque applied-research center, is trying to put on a robot's face.

The lab's BEGI prototype is a single-lens, passive 4D sensor built to mimic a compound insect eye, and it is one of the more explicit attempts so far to wire that frame-to-event transition into "physical AI", shorthand for AI that runs inside robots, drones, and other machines acting in the real world, not just on screens or in language models. BEGI's first public showing matters less as a product than as a map of how a research lab is trying to close the gap between biology and silicon.

The mechanism is the load-bearing claim. BEGI stacks a microlens array over a dynamic vision sensor (DVS), the same class of event-driven chip Prophesee pioneered. Each pixel fires only on local light changes, producing an asynchronous stream of "something moved here" events instead of a 60-fps image. On top of that, the microlens array shapes incoming light so the sensor can compute 3D depth from a single optic, the way an insect's compound eye turns many tiny lenses into one depth map without a second camera. The geometry is described in Ikerlan's program documentation as a single-lens alternative to a stereo camera pair, which means smaller optics, fewer calibration headaches, and lower data rates downstream.

That architecture fits the physical-AI use case directly. A robot chasing a moving target, a drone dodging a branch, or a manipulator grasping a tumbling object cannot afford the latency and power cost of pushing full 60-fps frames through a perception pipeline. It needs to know what changed, where, and how fast. Event-driven vision is built for exactly that question. A neuromorphic sensor, so named because it borrows the brain's "only fire when something happens" pattern, trades raw image fidelity for time resolution and energy efficiency, the two currencies embodied AI runs on.

The catch is in the silicon. BEGI's current prototype does not run on an Ikerlan-designed chip. It runs on the Prophesee IMX636, a 0.9-megapixel event-based sensor co-developed with Sony that is the closest thing the category has to commodity hardware. That is not a weakness in the prototype, it is the point. Ikerlan is using the best off-the-shelf event sensor it can buy to validate the optic and the system concept first, and treating the custom chip as a separate problem.

Project lead Xabier Iturbe told EE Times that Ikerlan plans to develop its own neuromorphic chip with Spain's CSIC and other Spanish partners over the longer term, but that work is on the roadmap, not the bench. For now, BEGI is a Prophesee-powered optic on a research timeline. Treating the roadmap chip as a near-term product would erase the difference between a published plan and shipped silicon.

That distinction is the only honest frame for the announcement. No third-party latency, power, or depth-accuracy numbers are available for the BEGI prototype in the public sources; the lab's own framing of "significantly improved data processing and power consumption" should be read as a direction, not a benchmark. The honest claim is mechanistic: a single-lens event-driven camera with microlens-derived depth is possible, Ikerlan has shown one, and it is aimed at physical AI. Whether that mechanism beats a stereo pair plus a frame sensor on a real robot is the experiment that has not been published.

The next milestone to watch is whether Ikerlan can move from the IMX636 to a custom die. Prophesee has spent years turning event vision into a buyable part. Ikerlan's bet is that an insect-eye optic plus a purpose-built neuromorphic chip is a category, not a feature. The bug is the marketing. The chip is the company.